Parallelisation only happens when `future` is attached than loaded

https://github.com/tidyverts/fabletools/blob/9794d225a37e79aaab18655692eca54bb3e1aed7/R/model.R#L100

future::plan(future::multisession)
rlang::is_attached("package:future")
#> [1] FALSE
library(future)
rlang::is_attached("package:future")
#> [1] TRUE

^{Created on 2022-06-27 by the reprex package (v2.0.1)}

Is this an intended behaviour? It took me some time to figure out why future::plan(future::multisession) does not work since the documentation does not require future to be attached. I thought it was something wrong with my future.

https://github.com/tidyverts/fabletools/blob/9794d225a37e79aaab18655692eca54bb3e1aed7/R/model.R#L27-L30

Wouldn't it be better to check if the future::plan is future::sequential when future is loaded rather than to check if it is attached? It would be nice to have the option to run it in parallel without ever putting future onto the search path.

Start with R/model.R at lines 27-30 and 100, then reproduce the reported behavior using future::plan(future::multisession) without attaching future. Determine how the current attachment check controls parallelisation; done means a loaded, namespace-qualified future plan enables parallelisation without requiring future on the search path.